[HLSL][Matrix] Add support for Matrix element and trunc Casts

farzonl · farzonl · commit 93183bce14f9 · 2025-11-25T11:21:01.000-05:00
fixes #168737 fixes #168755 This change fixes adds support for Matrix truncations via the ICK_HLSL_Matrix_Truncation enum. That ends up being most of the files changed. It also allows Matrix as an HLSL Elementwise cast as long as the cast does not perform a shape transformation ie 3x2 to 2x3. Tests for the new elementwise and truncation behavior were added. As well as sema tests to make sure we error n the shape transformation cast. I am punting right now on the ConstExpr Matrix support. That will need to be addressed later. Will file a seperate issue for that if reviewers agree it can wait.
diff --git a/clang/include/clang/AST/OperationKinds.def b/clang/include/clang/AST/OperationKinds.def
@@ -364,6 +364,9 @@ CAST_OPERATION(IntToOCLSampler)
 // Truncate a vector type by dropping elements from the end (HLSL only).
 CAST_OPERATION(HLSLVectorTruncation)
 
+// Truncate a matrix type by dropping elements from the end (HLSL only).
+CAST_OPERATION(HLSLMatrixTruncation)
+
 // Non-decaying array RValue cast (HLSL only).
 CAST_OPERATION(HLSLArrayRValue)
 
diff --git a/clang/include/clang/Sema/Overload.h b/clang/include/clang/Sema/Overload.h
@@ -198,6 +198,9 @@ class Sema;
     /// HLSL vector truncation.
     ICK_HLSL_Vector_Truncation,
 
+    /// HLSL Matrid truncation.
+    ICK_HLSL_Matrix_Truncation,
+
     /// HLSL non-decaying array rvalue cast.
     ICK_HLSL_Array_RValue,
 
diff --git a/clang/lib/AST/Expr.cpp b/clang/lib/AST/Expr.cpp
@@ -1934,6 +1934,7 @@ bool CastExpr::CastConsistency() const {
   case CK_FixedPointToBoolean:
   case CK_HLSLArrayRValue:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
   CheckNoBasePath:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
@@ -11773,6 +11773,10 @@ bool VectorExprEvaluator::VisitCastExpr(const CastExpr *E) {
       Elements.push_back(Val.getVectorElt(I));
     return Success(Elements, E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: support Expr Constant for Matrix Truncation
+    return Error(E);
+  }
   case CK_HLSLAggregateSplatCast: {
     APValue Val;
     QualType ValTy;
@@ -18163,6 +18167,10 @@ bool IntExprEvaluator::VisitCastExpr(const CastExpr *E) {
       return Error(E);
     return Success(Val.getVectorElt(0), E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: support Expr Constant for Matrix Truncation
+    return Error(E);
+  }
   case CK_HLSLElementwiseCast: {
     SmallVector<APValue> SrcVals;
     SmallVector<QualType> SrcTypes;
@@ -18756,6 +18764,10 @@ bool FloatExprEvaluator::VisitCastExpr(const CastExpr *E) {
       return Error(E);
     return Success(Val.getVectorElt(0), E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: support Expr Constant for Matrix Truncation
+    return Error(E);
+  }
   case CK_HLSLElementwiseCast: {
     SmallVector<APValue> SrcVals;
     SmallVector<QualType> SrcTypes;
@@ -18913,6 +18925,7 @@ bool ComplexExprEvaluator::VisitCastExpr(const CastExpr *E) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
     llvm_unreachable("invalid cast kind for complex value");
diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -188,6 +188,7 @@ Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr,
     case CK_HLSLArrayRValue:
     case CK_HLSLElementwiseCast:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_IntToOCLSampler:
     case CK_IntegralCast:
     case CK_IntegralComplexCast:
@@ -1290,6 +1291,7 @@ LValue CIRGenFunction::emitCastLValue(const CastExpr *e) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp b/clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp
@@ -534,6 +534,7 @@ mlir::Value ComplexExprEmitter::emitCast(CastKind ck, Expr *op,
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
@@ -1012,6 +1012,7 @@ class ConstExprEmitter
     case CK_MatrixCast:
     case CK_HLSLArrayRValue:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_HLSLElementwiseCast:
     case CK_HLSLAggregateSplatCast:
       return {};
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
@@ -5734,6 +5734,7 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprAgg.cpp b/clang/lib/CodeGen/CGExprAgg.cpp
@@ -1036,7 +1036,7 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
   case CK_ZeroToOCLOpaqueType:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
-
+  case CK_HLSLMatrixTruncation:
   case CK_IntToOCLSampler:
   case CK_FloatingToFixedPoint:
   case CK_FixedPointToFloating:
@@ -1550,6 +1550,7 @@ static bool castPreservesZero(const CastExpr *CE) {
   case CK_NonAtomicToAtomic:
   case CK_AtomicToNonAtomic:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
     return true;
diff --git a/clang/lib/CodeGen/CGExprComplex.cpp b/clang/lib/CodeGen/CGExprComplex.cpp
@@ -621,6 +621,7 @@ ComplexPairTy ComplexExprEmitter::EmitCast(CastKind CK, Expr *Op,
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprConstant.cpp b/clang/lib/CodeGen/CGExprConstant.cpp
@@ -1333,6 +1333,7 @@ class ConstExprEmitter
     case CK_ZeroToOCLOpaqueType:
     case CK_MatrixCast:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_HLSLArrayRValue:
     case CK_HLSLElementwiseCast:
     case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -2422,9 +2422,27 @@ static Value *EmitHLSLElementwiseCast(CodeGenFunction &CGF, LValue SrcVal,
     }
     return V;
   }
+  if (auto *MatTy = DestTy->getAs<ConstantMatrixType>()) {
+    assert(LoadList.size() >= MatTy->getNumElementsFlattened() &&
+           "Flattened type on RHS must have the same number or more elements "
+           "than vector on LHS.");
+    llvm::Value *V =
+        CGF.Builder.CreateLoad(CGF.CreateIRTemp(DestTy, "flatcast.tmp"));
+    // write to V.
+    for (unsigned I = 0, E = MatTy->getNumElementsFlattened(); I < E; I++) {
+      RValue RVal = CGF.EmitLoadOfLValue(LoadList[I], Loc);
+      assert(RVal.isScalar() &&
+             "All flattened source values should be scalars.");
+      llvm::Value *Cast =
+          CGF.EmitScalarConversion(RVal.getScalarVal(), LoadList[I].getType(),
+                                   MatTy->getElementType(), Loc);
+      V = CGF.Builder.CreateInsertElement(V, Cast, I);
+    }
+    return V;
+  }
   // if its a builtin just do an extract element or load.
   assert(DestTy->isBuiltinType() &&
-         "Destination type must be a vector or builtin type.");
+         "Destination type must be a vector, matrix, or builtin type.");
   RValue RVal = CGF.EmitLoadOfLValue(LoadList[0], Loc);
   assert(RVal.isScalar() && "All flattened source values should be scalars.");
   return CGF.EmitScalarConversion(RVal.getScalarVal(), LoadList[0].getType(),
@@ -2954,6 +2972,21 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
     llvm::Value *Zero = llvm::Constant::getNullValue(CGF.SizeTy);
     return Builder.CreateExtractElement(Vec, Zero, "cast.vtrunc");
   }
+  case CK_HLSLMatrixTruncation: {
+    assert((DestTy->isMatrixType() || DestTy->isBuiltinType()) &&
+           "Destination type must be a matrix or builtin type.");
+    Value *Mat = Visit(E);
+    if (auto *MatTy = DestTy->getAs<ConstantMatrixType>()) {
+      SmallVector<int> Mask;
+      unsigned NumElts = MatTy->getNumElementsFlattened();
+      for (unsigned I = 0; I != NumElts; ++I)
+        Mask.push_back(I);
+
+      return Builder.CreateShuffleVector(Mat, Mask, "trunc");
+    }
+    llvm::Value *Zero = llvm::Constant::getNullValue(CGF.SizeTy);
+    return Builder.CreateExtractElement(Mat, Zero, "cast.mtrunc");
+  }
   case CK_HLSLElementwiseCast: {
     RValue RV = CGF.EmitAnyExpr(E);
     SourceLocation Loc = CE->getExprLoc();
diff --git a/clang/lib/Sema/SemaExprCXX.cpp b/clang/lib/Sema/SemaExprCXX.cpp
@@ -5196,19 +5196,18 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
   case ICK_Incompatible_Pointer_Conversion:
   case ICK_HLSL_Array_RValue:
   case ICK_HLSL_Vector_Truncation:
+  case ICK_HLSL_Matrix_Truncation:
   case ICK_HLSL_Vector_Splat:
     llvm_unreachable("Improper second standard conversion");
   }
 
   if (SCS.Dimension != ICK_Identity) {
     // If SCS.Element is not ICK_Identity the To and From types must be HLSL
     // vectors or matrices.
-
-    // TODO: Support HLSL matrices.
-    assert((!From->getType()->isMatrixType() && !ToType->isMatrixType()) &&
-           "Dimension conversion for matrix types is not implemented yet.");
-    assert((ToType->isVectorType() || ToType->isBuiltinType()) &&
-           "Dimension conversion output must be vector or scalar type.");
+    assert(
+        (ToType->isVectorType() || ToType->isConstantMatrixType() ||
+         ToType->isBuiltinType()) &&
+        "Dimension conversion output must be vector, matrix, or scalar type.");
     switch (SCS.Dimension) {
     case ICK_HLSL_Vector_Splat: {
       // Vector splat from any arithmetic type to a vector.
@@ -5234,6 +5233,17 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
 
       break;
     }
+    case ICK_HLSL_Matrix_Truncation: {
+      auto *FromMat = From->getType()->castAs<ConstantMatrixType>();
+      QualType TruncTy = FromMat->getElementType();
+      if (auto *ToMat = ToType->getAs<ConstantMatrixType>())
+        TruncTy = Context.getConstantMatrixType(TruncTy, ToMat->getNumRows(),
+                                                ToMat->getNumColumns());
+      From = ImpCastExprToType(From, TruncTy, CK_HLSLMatrixTruncation,
+                               From->getValueKind())
+                 .get();
+      break;
+    }
     case ICK_Identity:
     default:
       llvm_unreachable("Improper element standard conversion");
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
@@ -3721,7 +3721,6 @@ bool SemaHLSL::CanPerformAggregateSplatCast(Expr *Src, QualType DestTy) {
 }
 
 // Can we perform an HLSL Elementwise cast?
-// TODO: update this code when matrices are added; see issue #88060
 bool SemaHLSL::CanPerformElementwiseCast(Expr *Src, QualType DestTy) {
 
   // Don't handle casts where LHS and RHS are any combination of scalar/vector
@@ -3734,6 +3733,10 @@ bool SemaHLSL::CanPerformElementwiseCast(Expr *Src, QualType DestTy) {
       (DestTy->isScalarType() || DestTy->isVectorType()))
     return false;
 
+  if (SrcTy->isConstantMatrixType() &&
+      (DestTy->isScalarType() || DestTy->isConstantMatrixType()))
+    return false;
+
   llvm::SmallVector<QualType> DestTypes;
   BuildFlattenedTypeList(DestTy, DestTypes);
   llvm::SmallVector<QualType> SrcTypes;
diff --git a/clang/lib/Sema/SemaOverload.cpp b/clang/lib/Sema/SemaOverload.cpp
@@ -162,6 +162,7 @@ ImplicitConversionRank clang::GetConversionRank(ImplicitConversionKind Kind) {
       ICR_C_Conversion_Extension,
       ICR_Conversion,
       ICR_HLSL_Dimension_Reduction,
+      ICR_HLSL_Dimension_Reduction,
       ICR_Conversion,
       ICR_HLSL_Scalar_Widening,
   };
@@ -224,6 +225,7 @@ static const char *GetImplicitConversionName(ImplicitConversionKind Kind) {
       "Incompatible pointer conversion",
       "Fixed point conversion",
       "HLSL vector truncation",
+      "HLSL matrix truncation",
       "Non-decaying array conversion",
       "HLSL vector splat",
   };
@@ -2060,9 +2062,10 @@ static bool IsFloatingPointConversion(Sema &S, QualType FromType,
   return true;
 }
 
-static bool IsVectorElementConversion(Sema &S, QualType FromType,
-                                      QualType ToType,
-                                      ImplicitConversionKind &ICK, Expr *From) {
+static bool IsVectorOrMatrixElementConversion(Sema &S, QualType FromType,
+                                              QualType ToType,
+                                              ImplicitConversionKind &ICK,
+                                              Expr *From) {
   if (S.Context.hasSameUnqualifiedType(FromType, ToType))
     return true;
 
@@ -2102,6 +2105,59 @@ static bool IsVectorElementConversion(Sema &S, QualType FromType,
   return false;
 }
 
+/// Determine whether the conversion from FromType to ToType is a valid
+/// matrix conversion.
+///
+/// \param ICK Will be set to the matrix conversion kind, if this is a matrix
+/// conversion.
+static bool IsMatrixConversion(Sema &S, QualType FromType, QualType ToType,
+                               ImplicitConversionKind &ICK,
+                               ImplicitConversionKind &ElConv, Expr *From,
+                               bool InOverloadResolution, bool CStyle) {
+  // The non HLSL Matrix conversion rules are not clear.
+  if (!S.getLangOpts().HLSL)
+    return false;
+
+  auto *ToMatrixType = ToType->getAs<ConstantMatrixType>();
+  auto *FromMatrixType = FromType->getAs<ConstantMatrixType>();
+
+  // If both arguments are vectors, handle possible vector truncation and
+  // element conversion.
+  if (ToMatrixType && FromMatrixType) {
+    unsigned FromCols = FromMatrixType->getNumColumns();
+    unsigned ToCols = ToMatrixType->getNumColumns();
+    if (FromCols < ToCols)
+      return false;
+
+    unsigned FromRows = FromMatrixType->getNumRows();
+    unsigned ToRows = ToMatrixType->getNumRows();
+    if (FromRows < ToRows)
+      return false;
+
+    unsigned FromElts = FromMatrixType->getNumElementsFlattened();
+    unsigned ToElts = ToMatrixType->getNumElementsFlattened();
+    if (FromElts == ToElts)
+      ElConv = ICK_Identity;
+    else
+      ElConv = ICK_HLSL_Matrix_Truncation;
+
+    QualType FromElTy = FromMatrixType->getElementType();
+    QualType ToElTy = ToMatrixType->getElementType();
+    if (S.Context.hasSameUnqualifiedType(FromElTy, ToElTy))
+      return true;
+    return IsVectorOrMatrixElementConversion(S, FromElTy, ToElTy, ICK, From);
+  }
+  if (FromMatrixType && !ToMatrixType) {
+    ElConv = ICK_HLSL_Matrix_Truncation;
+    QualType FromElTy = FromMatrixType->getElementType();
+    if (S.Context.hasSameUnqualifiedType(FromElTy, ToType))
+      return true;
+    return IsVectorOrMatrixElementConversion(S, FromElTy, ToType, ICK, From);
+  }
+
+  return false;
+}
+
 /// Determine whether the conversion from FromType to ToType is a valid
 /// vector conversion.
 ///
@@ -2141,14 +2197,14 @@ static bool IsVectorConversion(Sema &S, QualType FromType, QualType ToType,
       QualType ToElTy = ToExtType->getElementType();
       if (S.Context.hasSameUnqualifiedType(FromElTy, ToElTy))
         return true;
-      return IsVectorElementConversion(S, FromElTy, ToElTy, ICK, From);
+      return IsVectorOrMatrixElementConversion(S, FromElTy, ToElTy, ICK, From);
     }
     if (FromExtType && !ToExtType) {
       ElConv = ICK_HLSL_Vector_Truncation;
       QualType FromElTy = FromExtType->getElementType();
       if (S.Context.hasSameUnqualifiedType(FromElTy, ToType))
         return true;
-      return IsVectorElementConversion(S, FromElTy, ToType, ICK, From);
+      return IsVectorOrMatrixElementConversion(S, FromElTy, ToType, ICK, From);
     }
     // Fallthrough for the case where ToType is a vector and FromType is not.
   }
@@ -2175,7 +2231,8 @@ static bool IsVectorConversion(Sema &S, QualType FromType, QualType ToType,
       if (S.getLangOpts().HLSL) {
         ElConv = ICK_HLSL_Vector_Splat;
         QualType ToElTy = ToExtType->getElementType();
-        return IsVectorElementConversion(S, FromType, ToElTy, ICK, From);
+        return IsVectorOrMatrixElementConversion(S, FromType, ToElTy, ICK,
+                                                 From);
       }
       ICK = ICK_Vector_Splat;
       return true;
@@ -2474,6 +2531,11 @@ static bool IsStandardConversion(Sema &S, Expr* From, QualType ToType,
     SCS.Second = SecondICK;
     SCS.Dimension = DimensionICK;
     FromType = ToType.getUnqualifiedType();
+  } else if (IsMatrixConversion(S, FromType, ToType, SecondICK, DimensionICK,
+                                From, InOverloadResolution, CStyle)) {
+    SCS.Second = SecondICK;
+    SCS.Dimension = DimensionICK;
+    FromType = ToType.getUnqualifiedType();
   } else if (!S.getLangOpts().CPlusPlus &&
              S.Context.typesAreCompatible(ToType, FromType)) {
     // Compatible conversions (Clang extension for C function overloading)
@@ -6251,6 +6313,7 @@ static bool CheckConvertedConstantConversions(Sema &S,
   case ICK_Incompatible_Pointer_Conversion:
   case ICK_Fixed_Point_Conversion:
   case ICK_HLSL_Vector_Truncation:
+  case ICK_HLSL_Matrix_Truncation:
     return false;
 
   case ICK_Lvalue_To_Rvalue:
diff --git a/clang/lib/StaticAnalyzer/Core/ExprEngineC.cpp b/clang/lib/StaticAnalyzer/Core/ExprEngineC.cpp
@@ -560,6 +560,7 @@ void ExprEngine::VisitCast(const CastExpr *CastE, const Expr *Ex,
       case CK_VectorSplat:
       case CK_HLSLElementwiseCast:
       case CK_HLSLAggregateSplatCast:
+      case CK_HLSLMatrixTruncation:
       case CK_HLSLVectorTruncation: {
         QualType resultType = CastE->getType();
         if (CastE->isGLValue())
diff --git a/clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl
diff --git a/clang/test/CodeGenHLSL/BasicFeatures/MatrixTruncation.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/MatrixTruncation.hlsl
diff --git a/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixCastErrors.hlsl b/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixCastErrors.hlsl
